Abstract:
Machine and Tool Health Monitoring is in great focus nowadays to prevent machine and
tool breakdown penalty cost. It is essentially required to have a robust system that significantly
reduces machine downtime by monitoring machine and tool components during the machining
process and predict machine and tool health as a measure of preventive maintenance. Such system
will result in the reduced production cost and operator injury risks. This research proposes a novel
approach to monitor tool health of Computer Numeric Control (CNC) machine for a turning
process using airborne Acoustic Emission (AE) and Convolutional Neural Networks (CNN). Three
different work-pieces of Aluminum, Mild steel, and Teflon are used in this experimentation to
classify health of Carbide and HSS tools into three categories of new, average (used), and wornout tool. Acoustic signals from the machining process are studied in time and frequency domain
and it has been observed that in both domains, the features were weak to be utilized for effective
THM system development. Further, AE signal has been used to produce time-frequency based
visual spectrograms and then fed to a tri layered CNN architecture that has been carefully crafted
for high accuracies and faster trainings. CNN parameters fine-tuning trails have been carried out
by applying and observing different sizes and number of convolutional filters in different
combinations. CNN architecture with four filters in the first layer, each of size 5x5, gave best
results for all cases yielding 99.2% average classification accuracy. Retraining of CNN has been
done with initializing previous training learned weights that resulted in 99.6% accuracy with the
same number and size of filters. The proposed approach provides promising results for tool health
monitoring by robustly coping with environmental noise yet being cost effective and consistent in
performance for all the work-pieces and tool materials.
